P. Borrill

RDMA, the backbone of hyperscale data centers, suffers from a "completion fallacy" where data delivery guarantees don't ensure actual application-level data integration, leading to subtle but critical failures.

Forget assuming time flows forward: a new "Leibniz Bridge" framework reveals how time's direction emerges from entropy production and dissolves classic distributed systems impossibilities.

Cloud platforms silently deleting your files, phantom emails, and AI models with corrupted memories all stem from the same root cause: a failure to distinguish mere temporal succession from meaningful integration.

Datacenter networks are haunted by "ghosts"—topology knowledge failures due to link flaps that occur every 48 seconds at 2025 cluster scale—and existing mitigations are insufficient, but Open Atomic Ethernet offers a potential exorcism.

Ditch the "forward-in-time-only" assumption: a new Ethernet protocol uses round-trip validation to guarantee semantic integrity and achieve infinite consensus, blowing past the limits of RDMA and NVLink.

Atomic checkpoints in distributed AI/ML training are a myth: asynchronous systems with crashes make truly consistent snapshots a measure-zero event.

Ditch the clock: message passing's reliance on temporal order is a category error, and constraint satisfaction offers a more fundamental, symmetric alternative.

The "arrow of time" in computing isn't a law of physics, but a design choice baked into our protocols, meaning we can rethink decades of distributed systems theory.

The Unix filesystem's promise of atomic operations is a mirage, leading to widespread data corruption and costly failures across major cloud providers and databases.